0.9.69 is sooo sloooooooow

[Samuel Rydh]

On Wed, Apr 23, 2003 at 12:52:45AM -0700, Jason Simpson wrote:
> 
> I just recently did an 'apt-get update && apt-get upgrade' on my dual 604
> Debian unstable box, where it rev'd MoL to 0.9.69 and installed new kernel
> sources. I looked around on the net for what might have changed in the
> kernel to warrant new sources, and I got the impression that something was
> added to increase MoL stability on SMP machines. So, I recompiled my
> kernel with the new sources and recompiled the MoL modules, but mol.o kept
> complaining about not finding the 'hash_table_lock' symbol, which is what
> I thought had been updated.
> 
> I doublechecked the kernel sources and saw that 'hash_table_lock' had not
> actually been exposed, so I manually applied to patch to do so, recompiled
> the kernel again, and fired up MoL. It takes 15 minutes for MoL to boot
> MacOS 9.2 now.

Benh's kernel tree exports the hash_table_lock. Ideally, all
other trees should export it too...

> At idle, MoL is taking 95% of one CPU.
> 
> What's going on here? Has anyone else experienced such slowness? 0.9.68
> didn't have this trouble, and, actually, I never ran into any problems
> running it on this SMP machine to begin with. 

It sounds like you are running a slightly old pre-release version
of 0.9.69. A bug in the framebuffer tracking code causes this
slowdown (not only on SMP systems btw). The bug primarily affects
MOL in an X-window. The latest BK/rsync version of MOL does not
have this problem.

As for running old versions of MOL on SMP systems; it probably
works but it is not completely safe. The latest version also
reduces inter-cpu latencies.

/Samuel